Non-convexonlinelearningviaalgorithmic equivalence

Neural Information Processing Systems 

We study an algorithmic equivalence technique between non-convex gradient descent and convex mirror descent.